YOLOv10-SCI: single-chip microcomputer defect detection based on the improved YOLOv10 network

To address challenges in microcontroller defect detection—such as difficulties in recognizing small targets, interference from complex backgrounds, and low efficiency and adaptability of traditional methods—this study proposes a deep learning model, YOLOv10-SCI, based on an improved YOLOv10 architecture. The model integrates Space to Depth Convolution to enhance feature downsampling and reduce the loss of small defect features. It incorporates a Convolutional Block Attention Module to strengthen channel and spatial attention, and optimizes the loss function by introducing Shape Weights and Scale Adjustment Factors to improve bounding box regression accuracy. Results show that the enhanced YOLOv10-SCI achieves a Precision of 91.64%, Recall of 88.42%, F1-score of 90.00%, and mAP of 92.35%, representing improvements of 7.78%, 6.55%, 7.15%, and 6.06% respectively over the original YOLOv10, with only a marginal parameter increase (less than 1M). The YOLOv10-SCI network effectively reduces false positives and false negatives, providing a high-precision, high-efficiency solution for intelligent microcontroller defect detection, and holds significant value for improving quality and automation in electronics manufacturing.